One of difficulties in a tinnitus treatment is the absence of objective diagnosis. In order to solve this problem, in 2006, Tunner et al. proposed a method that causes a startle reflex with large pulse sound following tinnitus-liked background sound. This method compares the startle reflex magnitude according to the existence of gap region for a tinnitus diagnosis of animals. However, it is inappropriate to apply this paradigm to a human, because the tinnitus research for animal measures its magnitude of behavioral response. Therefore, this thesis suggests the analysis of auditory evoked response as the startle reflex of human using prepulse gap paradigm method. The idea is inspired by the most reliable hypothesis on a cause of tinnitus, which is abnormal activity of the auditory central nervous system, and by the fact that the auditory evoked response shares the same neural pathways with the acoustic startle reflex. This thesis presents the development of a system that measures auditory evoked response in accordance with prepulse gap paradigm method for an objective tinnitus diagnosis. The various parameters used in prepulse gap paradigm method can be configurable. Also, low noise analog and digital circuits are implemented for measuring very small size auditory evoked response, less than 1μV. Using the developed system, the auditory brainstem response and the auditory cortical response can be measured by simply adjusting gain and bandwidth of filters. This thesis also includes the developments of the adaptive weighted averaging algorithm to reduce the measurement time. In general auditory evoked response measurement, the acoustic stimuli are repeated from hundreds to thousands times and the measured signals are averaged to extract the auditory evoked response from the relatively large background noise. The prepulse gap paradigm method decreases the frequency of auditory stimulation by less than 1 Hz, resulting inefficiently long measurement time. This long measurement time causes discomfort with patients which could increase the noise in the signal. In this thesis, adaptive weighted algorithm was applied on the measured signal and the result demonstrated that measurement time can be reduced up to 70 percent compared to conventional ensemble averaging method.